Prompts, Not Keywords: How Buyers Actually Ask AI

How do prompts differ from keywords?

A keyword is something you'd type into a search box: short, fragment-like, stripped of context. "Best CRM," "payment gateway India," "accounting software freelancer." A prompt is closer to how you'd actually ask a person: a full sentence, usually with a constraint or two folded in — budget, team size, location, what you've already tried.

That difference isn't cosmetic. A search engine matches a keyword against an index and ranks pages. An AI engine reads the whole prompt, decides what it's really asking, and composes an answer from whatever it can retrieve — sometimes running several related searches behind the scenes to cover the question properly (Google calls this "query fan-out" for its own AI features; see what query fan-out means for content planning). The constraint in the prompt — "under $50," "for a 5-person team," "logged-out signup only" — changes which candidates even get considered, not just how they're ranked.

Practically: a page written to rank for the keyword "project management software" is optimized for a fragment. A page that actually answers "what's a good project management tool for a small remote team on a tight budget" is optimized for the question a real buyer asks. The second is a narrower target but a much more direct match.

Which intent types do buyers use?

There's no universal taxonomy, but these eight cover the prompts that show up in practice:

Illustrative prompts, written for this post — not pulled from a live scan.

Commercial, comparison and recommendation prompts matter most for visibility work, because they're the ones where an engine actually names and ranks brands. Informational and navigational prompts rarely produce a "best of" list — there's often nothing to be visible in.

How does funnel stage change the prompt?

The same buyer, researching the same category, asks differently depending on where they are:

Illustrative prompts, written for this post.

Awareness-stage prompts rarely mention a brand at all — they're testing whether an engine understands the category, not whether it knows you. Decision-stage prompts are where a stated constraint ("not a subscription," "under $50," "works with Shopify") does the actual filtering. If your product fits the constraint but isn't described anywhere that mentions it, the engine has no way to surface you at that moment, even if it named you earlier at the awareness stage.

What do price-framed prompts look like?

Price and budget constraints show up constantly in real prompts, and they change the answer, not just the framing:

• "What's the cheapest invoicing tool that still has recurring billing?"
• "Best project management software under $15 a user per month"
• "Free accounting software for a freelancer making under $50K a year"
• "Is there a one-time option instead of a monthly AI visibility subscription?"

A price-framed prompt forces the engine to filter, not just rank — tools that are good but don't fit the stated number tend to drop out of the answer entirely rather than just moving down a list. That's a different failure mode than "losing a keyword": the content can be accurate and well-written and still never surface, because nothing on the page states the number a buyer is filtering on. If your pricing page doesn't say the number in plain text, near the plan it applies to, an engine answering a price-framed prompt has nothing to match against.

How should this change research?

This isn't a niche concern. In an August 2026 survey of 171 search professionals across 36 countries, SEOFOMO found 75% now run a dedicated AI search strategy for at least some of their sites (up from 51% the year before), and 92% say they actively monitor AI visibility and citations in some form. The practice has moved fast. The research habits underneath it — keyword lists, volume-based prioritization — mostly haven't caught up yet.

Three practical shifts, in order:

1. Stop starting from a keyword list. Start from the real questions a buyer would type into a chat window — eight to twelve of them, covering several intent types and at least two funnel stages, not one "best X" phrase repeated with synonyms.
2. Write the constraint into the prompt, not just the category. "Best accounting software" and "cheapest accounting software that handles GST" will surface different answers from the same engine on the same day. Test both.
3. Check your own pages for the numbers buyers filter on. Price, team-size limits, region support, free-plan caps — if a price-framed or local prompt depends on a number, confirm that number is written in plain text somewhere an engine can retrieve it, not buried in a PDF or a sales call.

See how to build a prompt set that represents your buyers for a fuller process, and how AI decides which pages to cite for what happens after the prompt is asked.

FAQ

Is a prompt just a longer keyword? No. Length isn't the distinction — a prompt carries intent and constraints ("under $50," "for a 5-person team") that change which candidates an engine even considers, not just how it ranks them.

Do I need data on how often each intent type is used before I can act? No sourced, public breakdown of intent-type volume exists yet. Build a prompt set that covers the types your buyers actually use — informational through local — rather than waiting for a number that doesn't exist.

Which intent type matters most for brand visibility? Commercial, comparison and recommendation prompts are where brands get named and ranked. Informational and navigational prompts usually don't produce a brand list at all.

Do price-framed prompts need different content than regular ones? Mostly the same content, but the number itself has to be stated in plain text near the relevant plan or feature — not just implied or left for a pricing call.

How many prompts should a small team track to cover this properly? Eight to twelve, spanning at least three intent types and two funnel stages, repeated on more than one day — a single prompt, run once, tells you what happened once.

What to do next

1. Write eight to twelve real buyer prompts, not keyword fragments, covering at least three intent types and both an awareness-stage and a decision-stage question.
2. Add one price-framed and one local or constraint-based version of your main category prompt.
3. Check whether your own pages state the numbers (price, team size, region) those prompts filter on, in plain text.

ContextOwl sells one-time AI visibility reports. The AI Citation Snapshot is a free, single-question version of the same check.